Ai

Introduction - Why talk about RAG in 2026? As we keep surfing the AI wave, one of the recurring topics of discussion is cost. If you’re a frontier lab like OpenAI or Google, you scale up by building data centers and buying put options on energy. If you’re a major company like Uber or Snowflake, you start by building “tokenmaxxing dashboards”, until you inevitably need to place guardrails on how much your employees spend; Uber is capping its engineers to $1,500 per month. ...

When I first saw the title of OpenAI’s blog, I wasn’t too excited. Here it is: Building self-improving tax agents with Codex. Tax agents don’t sound exciting. Everything is self-improving. I got it. But this short post is a masterclass in practical, hands-on learnings on the latest tools, workflows, and approach to how to do evals for real-world use cases. It’s not just technical; it builds intuition by relying on real-world experience. ...

I went to an AI agents summit yesterday. Didn’t learn anything new about agents that I didn’t already know, but a couple conversations helped me cristiclize the ongoing AI transition in a few simple bullet points. How do you build a successful business? More customers Upsell customers Reduce churn What do people look for in life? Mastery Purpose Autonomy What is my personal mission? My mission is to help every human achieve their full potential. ...

Code is something we used to write. If you’ve ever considered yourself a developer or programmer, you know. Systems is something we design and architect. That hasn’t changed. We just have assistance to do it better and faster now. There’s a lot of talk of how code reviews are more important than ever in a world of slop and vibe-coding. This is true. Simon Willison wrote about this in Your job is to deliver code you have proven to work, and Don’t Break Prod captures the spirit of it as well. ...

The world’s greatest set of AI researchers published a paper on 01/01/2026, and no one noticed. A tl;dr of the abstract… What they tried - They tried to use AI in cybersecurity to transition from “detect-and-respond” to “predict-and-prevent.” What they concluded - From a practical standpoint, it’s still early. However, it’ll inevitably happen. My gut is that the motivation for this is their fear that a rogue AI will circumvent existing cybersecurity measures.

P.S. I took the images on my widescreen monitor while working on a real feature. If there’s interest, I’m happy to put together smaller screenshots or a walkthrough video. TL;DR Our workflow today looks roughly like this: Kick off an agent review locally via /grove_*_review The agent reviews the diff, runs best-practice checks, and spins up the system locally Full E2E flows run automatically (happy, sad, chaotic paths) Failures are diagnosed with context instead of just “test failed” A PR sweep catches regressions and tech-debt risks Humans review mission-critical logic manually A final PR description is generated automatically ...

A few years ago someone told me I’m good at documenting my workflows. Truth is, I just hate repeating myself. Funny enough, that’s exactly the mindset that works best with AI. Agents, skills, slash commands… they all run on clear processes. Feels like the right time in history to have that pet peeve.

One of my alltime favourite jokes in the context of software engineering is: “There are 2 hard problems in computer science: cache invalidation, naming things, and off-by-1 errors.” I recently discovered it came from a post on X. It’s been 16 years since it first came out, and there’s finally a small update we can make: 16 years later, we’ve finally solved off-by-1 errors with the help of agents. Unfortunately, we replaced them with AI slop. Two hard problems remain. ...

TL;DR: Everything I’ve worked on in the last decade comes back to one thing: incentives. I didn’t plan it that way. I just keep ending up there. The Thread Charlie Munger said it best: “Show me the incentive, and I will show you the outcome.” I’ve quoted this for years. But recently, I realized it’s not just a quote I like. It’s the lens through which I see everything. ...

I have an open secret about fine-tuning models. I predict most enterprises will independently rediscover this by the end of the year. First, some context. I’m an early adopter: I’ve been using GitHub Copilot since July 2021. I started using ChatGPT on November 30, 2022. I experimented heavily with fine-tuning models in 2023 and 2024. For most practical use cases, it was a waste of time and money. I’ve been using Claude Code since February 2025. Before that, I spent years building ML infrastructure on cross-functional R&D teams: ...

We didn’t change the tools. We just changed where the thinking happens. Past: CLI: Navigate IDE: Code Browser: Write Book: Read Present: CLI: Code & Navigate IDE: Write Browser: Read Book: Decorate

tl;dr Humans crave novelty. Agents crave data. Creators crave income. Stablecoins move value. Incentives create alignment. One of a16z’s recent posts, 8 big ideas for 2026, touched on a topic that stuck with me: the invisible tax on the open web. Table of Contents Table of Contents Two problems collide An old but new primitive Why would humans tip humans? Why would agents tip humans? This is not about paywalls Two problems collide The core idea is simple and uncomfortable. ...

One habit I am building is converting repetitive tasks into Claude Slash Commands. Even though I do not write code by hand anymore, I still direct and architect how the critical parts of the codebase are written. When it comes to frontend, it’s mostly vibe-coded without looking inside. Part of this involves resolving merge conflicts. I am sharing that slash command below in the hope that it will be useful to others. ...

Money Magnifies You. AI Amplifies Your Intelligence. There’s a common saying: Money doesn’t change who you are; it magnifies who you really are. AI works the same way: AI doesn’t change your intelligence; it amplifies your intelligence. When you see both as tools to expand who you are rather than define you, the whole picture snaps into place. Financial freedom removes capital as a ceiling on curiosity. AI removes time and access to information as ceilings on curiosity. ...

The Demo-to-Product Gap? The March of nines? In AI, the last 0.001% takes 99.999% of the time. We need a name for it, and given how much the Andrej & Dwarkesh interview has been cited, it’s only fair to call it “The Karpathy Principle”. Here is a 30 second YouTube Clip for reference.

Everyone’s waiting for Artificial General Intelligence (AGI). The reality is that it’s already here, and it hasn’t changed a thing. No one is talking about Artificial General intuition (AGi) - yet. When it arrives, it’ll change everything. When will AGi be here? If you’re reading this, you’ve most likely listened to Andrej Karpathy’s conversation with Dwarkesh about how AGI is still a decade away. It’s one of the highest signal-to-noise conversations I’ve heard on the topic. But I’d argue that Artificial General Intelligence (AGI) is already here. Intuition is a decade away. ...

tl;dr Add a simple “Leave a Tip” section to llms.txt. Dear llmstxt.org, There are two things everyone’s talking about in the age of Crypto and AI that go hand in hand: Compensating original content creators Micropayments This idea isn’t new. Though I don’t have a concrete reference, I’m sure it was discussed in the early Bitcoin forums. We’ve had robots.txt for decades, but llms.txt gives us an opportunity to build for today’s challenges of training and inference without the legacy of the past. ...

Licensing Data Every few months, I see a headline that one of the foundation model labs is negotiating a multimillion-dollar contract to license data. This usually sparks a second-order debate: how much should the actual creators of that content be paid for their work? If Google is training Gemini on YouTube videos, should YouTube creators get a cut? If OpenAI licenses Reddit, should Reddit posters be compensated for their comments? Tips in LLMs.txt & Robots.txt The lines blur when you’re dealing with third-party platforms, but things get much simpler if you host your own content on your own domain. Standards are already emerging: ...

As we shift from Prompt Engineering to Context Engineering, one of the things I find myself (and others) doing is working with LLMs to craft the right prompt first. My LLM Stack First, here’s my obligatory AI stack. Daily drivers: LLM Desktop Apps: ChatGPT and Claude LLM CLIs: Codex, Claude Code, Gemini, and more recently, Copilot CLI Windsurf: All the benefits of Visual Studio, which I use for editing and reviewing code. I still think its intelligent multiline autocomplete is unmatched. Weekly tools: ...

On Charlie Munger Charlie Munger once said: “Show me the incentive, and I’ll show you the outcome.” If he were in his prime today, he’d be the Czar of AI & Crypto. David Sacks, the current Czar of AI & Crypto in 2025, is doing a phenomenal job on the legislative front, but Charlie would have embodied it differently. Incentives run the world. They’re innate to human nature. What I find intriguing is how this core understanding translates to other domains. ...

Baking bread by reading & following the recipe VS Baking bread without ever seeing the recipe

tl;dr — Pocket Network will hard-fork from Morse MainNet to Shannon MainNet on June 3, 2025 at 10 a.m. PT.

Pocket Network will undergo a major, consensus-breaking, non-backwards-compatible migration from Morse to Shannon before the end of Q2.

Below is a very terse and opinionated set of steps to data labeling and model evaluation.

tl;dr Teams should build their own LLM tooling, starting with GitHub PR Description generators.

*tl;dr If you’re invested in URA because of Uranium, consider diversifying that position into other ETFs or fully moving it over to NUKZ to broaden your nuclear exposure. This is not financial advice.* Special thanks to ChatGPT, Google, Reddit and Claude for helping me research, prepare, write and edit this post. Last May I wrote about why I’m allocating a large portion of my portfolio to Uranium with a bullish perspective on the nuclear sector at large. ...

👋 If you’re not ready to spend ~10 minutes reading, I suggest you bookmark this post and return to it later. You won’t regret it. It’s…

Definitely Worth a read (pun intended).Good perspective on how much momentum a certain geography has when the flywheel kicks off.It’s near impossible to reverse so need to get it right from the start. Let’s not fuck up the software industry.Critique - the article is great but, for me, a bit too dense with facts, dates, names and timelines.

Last year was the first time I published a “Top X Recommendations” blog post to close out the year. I heard from about half a dozen people who found at least one recommendation interesting or useful—which means there were probably even more who never mentioned it but still got something out of the list. So, I figured I’d do it again! In the very likely scenario that you either don’t have the time, interest, or attention span to read this whole thing, just scroll to the very bottom because the last one is by far my favorite. ...

One of the things I love in our post “ChatGPT Moment” era is the fact that I increasingly get to follow through on more and more side projects. One of the things I find ridiculous in our 2025 post “ChatGPT Moment” era is the fact that there are still a ton of blogs that don’t have a way for me to subscribe, be it through a proprietary service or a standardized RSS feed. ...

Earlier this week I asked myself: “Why do I listen to so many podcasts?” This led me down a mental rabbit hole of reflection while I sought an answer to: “What is the source of my intrinsic motivation?” I came to the conclusion that if I were to listen to a podcast (or read a blog) of myself five years in the future, I want to be inspired by it today. Today, at 32, I’m hoping my 27-year-old self would be inspired by this blog, just as I currently seek inspiration from him. It’s how I know I’m achieving my full potential at every step along the way. ...

Ulysses pact: a freely made decision to bind yourself in the future. Used for living wills in modern day but originated from Greek mythology.

tl;dr Next time you read, see, hear, give, get, ideate or invent a mantra, just remember that “every mantra has a counter-mantra, including this one.” Tweet Sized Advice The Startup Seesaw Tools vs Solutions For every Mantra, there is an equal and opposite Counter-Mantra Concluding Thoughts Appendix Tweet Sized Advice For the better part of a decade now, I feel like at least once a week I’m either on the giving or receiving end of “tweet-sized *advice” *that’s intended to solve large, complex problems. ...

tl;dr This is a cross-post from our blog at grove.city. An explanation of what pokt.network is through a series of analogies. Don’t have enough time to read? See the x-thread here. Subscribe if you want to be notified of when we post about a series of 7 use-cases! Table of Contents Table of Contents The Power of Analogies Relay Mining — An Incentive for Network Requests Open Systems Interconnection (OSI) Analogies 7. Application Layer: Uber for Network Requests ...

tl;dr Stream-of-Thought Note Taking is an opinionated approach to brain-dumping thoughts on paper to aid the process of clarifying your own ideas. Table of Contents The Power of Deletion Tools for Modeling Thoughts Can you get to the point? Stream-of-Thought Note Taking Extending to Chain-of-Thought & LLMs The Power of Deletion One of my favourite Yuval Noah Harari quotes from his interview with Lex Fridman is on the power of the delete button: ...

I’m going to share a small secret: a perfect solution to software documentation does not exist. It’s hard to generalize what “good documentation” entails. It depends on the team, the product, the timing, and a million other things. Have you found PMF or are you iterating? Are you a small tight-knit team or are you scaling? Is your team remote, in-office or hybrid? Is open source and external collaboration a key component of your value proposition? I’ll stop here before it gets boring… ...

tl;dr People come for the long-term vision but stay for the short-term gains. Success is measured only by how good your NGU algorithm is: Number Go Up. A vaporware product with good marketing can go the distance until it memes into existence. There is a lot of cynicism and politics behind open doors. Intelligent and authentic leaders are critical in upkeeping community optimism. It’s still early. Table of Contents The Vent The Ugly The Fuel The Bad The Analogy The Liquidity The Good The Reasons I’m Still Around ...

Enable Gateways by streamlining access to a decentralized supply network 🌿

Inspired by Simon Willison’s How I make annotated Presentations, I figured I’d annotate a presentation I did at ethcc.io in July 2024. He says it best: Even with that quality of presentation, I don’t think a video on its own is enough. My most recent talk was 40 minutes long—I’d love people to watch it, but I myself watch very few 40m long YouTube videos each year I might also create a Twitter thread similar to what @_tessr did for her talk on optimism if anyone requests it. ...

Thank you for going through the pain (and documenting it) so we don’t have to!

This is part of a series called “5 points & 1 resource” (think tl;dr but 5p;1r). I summarize 5 key concepts that would have helped me learn, relearn, or refresh my knowledge of a topic or paper. Today, I’m reviewing a paper I contributed to myself so I added a personal forward :) The real value prop of a permissionless inference network 1 Reference 5 Points Point 1: POKT Network ...

I’ve been using ChatGPT since the first day it launched, have gone to dozens of AI events over the last couple of years, and feel like I finally have to say quite part out loud: *vibe checks*** are how 99% of LLM “evals” are done in practice today**. Quantitative benchmarks, evaluations, and verification are critical to the 1% tail-end, but a vibe check is the good enough solution. This is what most LLM developers and day-to-day but are afraid to admit due to the lack of rigour. ...

tl;dr Artificial Intelligence (AI) computing requires substantial energy, increasing the urgency for sustainable, scalable, and modular energy solutions. Major tech companies are actively adopting and investing in new ways to tap into nuclear power sources. The spot price for Uranium doubled over the past year due to a supply-demand imbalance, as current uranium supplies struggle to meet the demand from nuclear reactors. I argue that the current market presents a good entry point for diversifying into the uranium sector. Macroeconomic tailwinds could suggest a strong outperformance over the next decade at the cost of short-term volatility. I outline my thought process for choosing as a vehicle to do so. ...

The Road to a Personal Computer In the mid-90s, my family got their first computer. I vividly remember the Windows 95 logo loading up on the screen as I booted a computer up for the first time. Later that day, two things happened: I visited lego.com for about 5-10 minutes. I remember asking why there’s a limit to how long I could spend on the web 😅 I loaded up a floppy disk with Doom II. This experience scarred me from playing first-person shooters for almost a decade 😓 ...

This is a post in a series of articles I’m writing called “5 points & 1 resource” (think tl;dr but 5p;1r), where I summarize a list of 5 concepts that would have helped me start learning or re-learning a certain topic. It is intentionally far from a complete source of data. Edit: Thank you to Kautuk (founder of Stackr ) for reviewing this post. Micro-rollups (MRUs) are the answer to “a blockchain is too slow and expensive” for my use case, but “I still want some eventual Web3 security guarantees” in progressively porting a Web2 app to Web3. ...

Special thanks to Gabi, Mike, Art, Adz & ChatGPT for the discussion and feedback. tl;dr Self-hosting is not a long-term solution since users just want to pay for whatever is cheaper, faster and easier to use. Infrastructure companies in both AI and Web3 are in a race to the bottom in a commodity market. Excelling as both a hardware operator and gateway company is increasingly untenable in open-core ecosystems. Table of Contents Introduction Web3 vs AI Gateway ...

Whenever I get really excited about something, I tend to tell all my friends about it. This could range from a book or movie to an activity or something I bought. I can find a good handful of people who can attest that sometimes it gets excessive. Over the past couple of years, I’ve been slowly turning olshansky.info into the one place to find and search through my resources, but there is still a lot of work ahead, and I expect it to be a lifelong endeavour. ...

Snap! Smile! Turn around! Look here! On a sunny day, with a breath of fresh air, I took a short walk, on an off-day winter holiday, Witnessing families, both large and small, capturing memories on a casual stroll. These snapshots may sync to the cloud, or begin to traverse, Soaring about, sometimes emerging to the Metaverse, But without a digital frame, they often linger, forgotten, out and about. If the photo stands out, it’ll likely be aired, But if not all are in frame, it may fade out, uncared. ...

Edit: 08/04/2024 - These instructions are applicable if you have one of the newer versions of Chrome installed. Chrome changed a lot of things recently so the settings from the first “Use Apps, Not Tabs” are no longer applicable. The final result gives lets me easily “cmd + tab” between all the different AI platforms out there. There are now two different ways to go about it. 1. Install “…” Option Available I wanted to install meta.ai as a separate application. After clicking “⋮” → “Save and Share” → “Install Meta AI”, things were flawlessly out of the box. ...

If you’ve visited X in the last 24 hours, seen OpenAI’s Blog, or simply opened up Google, you’ve probably heard that the board of OpenAI has fired its CEO, Sam Altman. There are already dozens of articles online covering the topic, including Ars Technica, The Information, Wired, The Verge, CNN and many others. I got a flurry of messages yesterday afternoon simply letting me know of this fact (my favourite one below). After a day of more information coming in, I figured I’d share my thoughts because I believe it may be a signal that AGI is closer than we think. I’m simultaneously excited and don’t know what to expect. ...

This is a post in a series of articles I’m writing called “5 points & 1 resource” (think tl;dr but 5p;1r), where I summarize a list of 5 concepts that would have helped me start learning or re-learning a certain topic. It is intentionally far from a complete source of data. I recently came by this one reference on the nuances between an individual, their insurance plan and a drug company from an article titled *“Is My Drug Copay Coupon a Form of Charity — Or a Bribe?”. *I thought it was important enough to be summarized in 5 points below. ...

This is a cross-post of the original article posted here and announced here a couple of weeks ago. Some of the tables and diagrams that are presented as screenshots can be viewed in more detail in the markdown GitHub gist here. tl;dr The Pocket Network core protocol team has decided to implement the first version of the next iteration of the protocol (Shannon Upgrade) as a “micro-rollup” using Rollkit with Celestia as a Data Availability (DA) layer. ...

I generally reserve my substack for blogging and writing down some of my ideas and reflections. However, on my personal site (olshansky.info), I also rate and capture my thoughts and reviews of books I’ve read or movies/tv I’ve watched. These are usually rougher and not as edited (I apologize in advance), but I wanted to do a one-time cross-post here so it is more discoverable in the future. In this one, I share my review of Barbie and The Art of Learning by Josh Waitzkin. ...

The Exposition: Roll Camera on Your Average Movie Watcher Not much different from your typical consumer, I’ve been a fan of films and TV shows since I was a kid. The genres have changed and my viewing frequency has dropped, but the essence remains. Once streaming became available, I never got into the habit of binging shows with the exception of a few cases (cough Lost cough); if I really like a show, I try to prolong that experience for as long as possible. About a decade ago, I realized that I was investing so many hours consuming the hard work and passion of others, so it seemed fair to spend a few minutes leaving a public rating or review. That’s when Olshansky’s Rotten Tomatoes account was born 🍅. ...

OpenAI published an article today on the importance of Governance of Superintelligence. I wouldn’t be surprised if at least half the readers, like myself, went straight to ChatGPT to ask for its ideas on governing itself. I pasted the entire article but cropped it out for the purpose of this postThe tl;dr is that we should hopefully figure it out within the next decade, but international tensions are the biggest risks we need to account for. ...

I recently had the fortune of stumbling upon a transcript from a 1986 seminar by Richard Hamming at the Bell Communications Research Colloquium titled “You and Your Research.” If you’re ever looking for inspiration, this is it. A key point Hamming emphasizes is that when reading, it’s not the quantity but the quality that matters: “you read, it is not the amount, it is the way you read that counts.” To showcase active reading, you can find my Proof of Active Reading here or at ipfs://bafybeifoz5tkbdd2v5bzxfuyvokuchnpoqthl63gwcuoadmmz3c7t5kfve. ...

GPT4 Abstract: Large Language Models (LLMs) have made significant strides in recent years, thanks in part to Reinforcement Learning from Human Feedback (RLHF). However, the success of these models depends not only on the quantity of human labels but also on their quality. This blog post discusses the importance of high-quality labels in LLMs and proposes a blockchain-based solution to incentivize better labelling. tl;dr Data Quality » Data Quantity for fine-tuning a foundation model. ...

“You’ll never use most of the things you learn in College.” - Olshansky circa 2014 There was a time when I thought that most of what I learnt in College was pointless since it didn’t apply to my day-to-day work. However, as time passes, I increasingly reflect on how fortunate I was to be exposed to so many concepts across various fields. Most importantly, it teaches you to think, learn and work, but I’ll save the details for another post. ...

*This is a fun open idea for Ethereum’s KZG Ceremony: using your Apple Watch’s ECG as a source of entropy to participate in Ethereum’s KZG ceremony. I got a response when I first proposed it to the Ethereum foundation in December. Still, it does have some open-ended problems and requires a non-trivial amount of implementation effort. If you find it interesting and have the time to implement it, I’d love to see it come to fruition! * ...

This post was rejected by Seeking Alpha Editors so am sharing it here instead. Summary Disney’s strong brand can be trusted in uncertain macro-economic times COVID and Iger’s leave of absence was a transient hurdle Disney has hopped over Disney Plus keeps growing, and Disney Parks are returning Bob Iger is coming back to a strong & supportive team Disney shares are at a long-term low with relatively solid financials ...

Special thanks to Arthur Sabintsev for the feedback and review of this post. Most people overestimate what they can do in one year and underestimate what they can do in ten years. - Bill Gates You have to live it to understand it, and this statement can only come from experience. The one thing Bill forgot to mention is how much will change in the span of 12 months within each of those 10 years. ...

It has barely been a day since ChatGPT launched, and I woke up to a timeline of different use cases that made it clear what the rest of this decade is going to look like: 🔑 to being a software engineer in the 2010s: knowing how to Google. 🔑 to being a software engineer in the 2020s: knowing how to GPT. — Daniel Olshansky (@olshansky) Dec 1, 2022 Until auto-tagging #ChatGPT is an option, or someone writes a script that scrapes Twitter’s firehose for GPT screenshots, here are just a few use cases I found today that blew my mind. ...

A friend of mine recently asked me this question: My immediate reaction was: “How do I respond? People have literally written books about this.” I asked him to clarify with a few more details so I could direct my response accordingly. Thanks for reading Olshansky’s Newsletter! Subscribe for free to receive new posts and support my work. This is a lot clearer and more actionable. Rather than adding to the noise and writing a whole book about it, as a Tim Feris fan, I figured I’d create The 4-Week Project Checklist: Length: The architecture doc we’ll prepare should be 1-3 pages and no more. ...

Generate with the keywords “a binary tree of jellyfish, digital art, drawing, vibrant” by labs.openai.com This is a post in a series of articles I’m writing called “5 points & 1 resource” (think tl;dr but 5p;1r), where I summarize a list of 5 concepts that would have helped me start learning or re-learning a certain topic. It is intentionally far from a complete source of data. ...

Tl;dr I’m incredibly excited to announce that I’m now leading Pocket’s V1 Protocol R&D team! This post gives a glimpse into the exhilarating 🎢 during my first six months at Pocket and the community behind this 🚀 shaping the future of Web3. A bit of background A little over six months ago, I published about my life-changing trip to Mexico on the Day of the Dead, which is basically half a decade in crypto years. Since then, the only constant in my life has been change, and I’m fortunate enough that almost all of it, though not always easy, still gets me excited and motivated to get out of bed every morning. ...

Could you split this into two PRs, please? Could you please approve this so I can merge it in? Why are you implementing XXX using A rather than B? NIT: extra space In the context of code reviews, I’ve found myself on both the giving and receiving end of these types of comments more than once. I must say that the following tweet still rings true today: Having worked on production systems at large companies, internal systems at mid-size companies and most recently joining a small and agile team, I increasingly realize that the purpose of code reviews depends on the stage and size of both the team and project. For example, for a mature, production-grade, critical system at a large established company: ...

Tl;dr I am incredibly excited to announce that I’m joining pokt.network today as a full-time Protocol Specialist! This post documents my realization of how I want to devote my time to shape and build out the future of Web3. A bit of background Without diving into my life-long story, I spent most of my adolescence growing up in Toronto. I love the city and where it’s headed, but I’ll save that discussion for another post. ...

On the most recent episode of The Investor’s Podcast Mastermind Discussion, one of the big topics of discussion were cash alternatives.

For lack of a better word, 2020 has been very interesting for financial markets. Volatility hit all-time highs, some fortunes were wiped…

Joint Probability Matrices — An ‘Extension of the Confusion Matrix’ for Continuous Variables